Ontological positions

When working with scientists or researchers from different domains, it is very important to keep in mind that we do not necessarily have the same expectations from our scientific work and do not necessarily rely on the same concepts to assert causal relationships. Taking a transversal approach towards a given problem therefore requires members to recognise the role, strength and limitations of each scientific domain and to clearly understand how they contribute to solving societal problems.

Basic science usually considers causality as the physical mechanism that explains the link between cause and effect. It aims for a better understanding of complex interactions in matter (mechanical approach). For epidemiologists, causality is seen as the link by which we can modify systems. This more practical approach, therefore,

PART  2  –  Causation  in  traffic  security  research  

does not seek to explain causes but only to identify them (probabilistic approach). For example, cannabis smoking leads to an increased risk of motor vehicle collisions.⁶ In psychology, researchers mainly aim to model and explain behaviour through mental states (causal modelling). Mini Mental State Examination used to detect cognitive disorders is an example of causal modelling.⁷ Sociologists mainly use statistics to test models associating social indicators to group behaviour (statistical inference). For example, traffic injuries are more frequent for children from lower social positions.⁸ Finally, causality in the legal system is founded in the concept of conformity. Moral entities are held responsible for an event if their behaviour has been non-conforming and it can be proven that the event was caused by this behaviour (non-conformity causality). Contraflow driving on the expressway, for instance, would be considered as a cause of an accident, if it were to occur.

2.3.1 Mechanical causality and basic science

Mechanical causality in behavioural science is the explanation of how the parts of a system give certain properties to a whole system through the interactions between parts.⁹ Within a system, structural modifications of entities (effects) are produced by activities (causes).¹⁰ The physical exchanges that take place show regularity ¹¹ and are therefore replicable in similar conditions (deterministic approach). When studying one to one links between cause and effect, for a condition to be a cause of an effect the presence of this condition must lead to the effect and the effect must not occur without it.¹² This strict determinism approach is the milestone of basic science and is often referred to as sufficiency causation or necessity causation. In neurobiology and behavioural neuroscience, this approach of causality is widely used to identify and describe underlying biological mechanisms at a molecular level.¹⁰

In our fictional scenario (Point 2.1), the fracture will depend on the propensity of the limb to deform and resist to the deceleration due to the impact of both moving objects. The mass of the car, intrinsic properties of bone not to withstand torsion and the trajectory of the kick scooter on the sidewalk are therefore some of the mechanical causes of the fracture. Mechanical causality can also highlight the mechanism by which the medication acts on the cardiovascular system.

Advantages: In behavioural science, mechanical causality has many advantages over other definitions of causality. First, it can distinguish causes from consequences.

Secondly, it explains underlying mechanisms that help consider possible solutions to prevent an event from taking place. Thirdly, the causal link is verifiable using an experimental approach in a controlled environment.¹³ This approach is therefore epistemologically sound and provides a causal explanation we can rely on.

Limitations: Mechanical causality has some limitations. First, mechanical causality does not provide any information on the likelihood of a situation occurring and therefore cannot attribute different weight to different causes implied in a phenomenon. In other words, at a mechanical level, all causal factors are equally responsible for a given event. Secondly, defining mechanical causality highly depends upon the amount of control that can be exercised over the environment in which the studied phenomenon takes place. When studying complex phenomena, mechanical causality can therefore often arise in experimental or hypothetical observations that are not relatable to real life situations. Finally, mechanical causality is limited in determining states of systems through time. Discoveries in quantum physics have revealed that the laws that govern matter are not as deterministic as we presumed them to be. At a sub-molecular level, some forces that govern the organisation of

matter are, by their nature, unpredictable.¹⁴ These forces provide the scientific explanation of the part of unpredictability that can be observed at other levels (molecular, biochemical, cellular, intercellular, organism, inter-organism, environment etc.) and are called stochastic events. In behavioural science, stochastic events at a quantal level explain variations in gene transmission,¹⁵ in gene transcription,¹⁶ and in variability in synaptic transmission and plasticity.^{17, 18} When studying human behaviour, stochastic events are therefore not incompatible with mechanical causality, but they prevent this approach from predicting the state of an uncontrolled system for more than a few milliseconds.

2.3.2 Interlevel constraints

Systems can be defined at different levels of organisation (Figure 1). When studying behaviour in neuroscience, a common procedure is to enhance a structural change to a system at a molecular level, and then observe the effects on individuals’ behaviour (bottom-up experiments).

Figure 1: Levels of organisation in biology and behavioural science

An example of a bottom-up experience is to generate conditional knockout mice and assess the role of a specific protein in a specific brain region upon a behaviour. It is however also possible to have individuals undergo different tasks (behaviour level) and then compare effects on structural changes at a molecular level (top-down experiments). Exposing subjects to a paradigm and using functional magnetic resonance imaging to assess topographic vascular changes is an example of a top-down experiment. This presumes we accept that mechanical causality can affect different levels of organisation. By reduction, changes at any level should theoretically be explainable by interactions between elementary particles. However, at this level, mechanical causality cannot be used anymore; there are no more parts to which to reduce an explanation. We can therefore admit that mechanical causality is designed to explain interactions between parts of a system but not to reduce the

PART  2  –  Causation  in  traffic  security  research  

explanation to its lowest level. Kistler ⁹ overcame the problem of interlevel causality and mechanical causality by distinguishing constraints from causality. For him, “a constraint limits the possibilities of evolution or change accessible to a system” and this takes effect within and across levels. Therefore, mental causality can be seen as the result of constraints rising from neuronal activity.¹⁹ Admitting interlevel constraints has major implications in the conception of mechanical causality in behavioural science. When studying mechanical causality across different levels, it is necessary to provide physical explanations of how changes at one level affect conditions at another level to the point that it constrains change in this system.

In our fictional situation, interlevel constraints and mechanical causality could provide explanations for the neurobiological mechanisms taking place in dementia that led to the driver’s loss of memory. It could also explain the underlying biological constraints due to the physician’s feelings of anxiety and the consequent effects on attention shift. This could provide explanations as to why the physician did not think of exploring his patient’s cognitive state.

Advantages: First, interlevel constraint widens the field of mechanical causality to include causality by omission and non-occurrence causality. The direct physical link between the absence of a cause and its effect does not have to exist anymore. In other words, it is conceptually sound to consider that the absence of a cause at one level can modify constraints at another level. Public health is often concerned about how to prevent something from happening. When developing treatments, interlevel constraint is therefore indispensable to link the mechanical effect of the treatment with the non-occurrence of symptoms.²⁰ Secondly, interlevel constraints are essential for mechanical causality to provide explanations on how changes at a molecular level can enhance changes at a behavioural level and vice versa (Figure 2). This widens the field of application and supports treatments other than those at the molecular level.

Finally, the concept of interlevel constraints opens the field of personalised medicine.

From both symptoms and biological markers, treatments can be adapted to individuals to prevent adverse effects from occurring.

Figure 2: Interlevel constraints linking environment,

Limitations: This approach nevertheless does not offer the possibility of accepting links between levels without acknowledging the physical constraints that one level has over another. Given the difficulty of providing such explanations at higher levels, we can use epidemiology, psychology and sociology to overcome this problem, by using a probabilistic approach to define causality.

2.3.3 Probabilistic causality and epidemiology

Science might seek to know the extent to which entities will tend to interact with each other and participate in an effect, without wanting to provide any explanation of underlying mechanisms. In this situation, a probabilistic approach instead of a mechanical approach of causality is often preferred. Even if mechanical causality can take multiple factors into consideration (sufficient-component causes),²¹ this approach fails to take interactions into consideration or correctly model dose-effect response.

Probabilistic causality in epidemiology has made it possible to attribute probability of multicausality, evaluate interactions among causes, and to estimate the strength of causes and the attributable fraction of a cause to a disease.²² In observational studies, it is, however, often impossible to separate causal from non-causal associations. To limit false assumptions, Hill ²³ proposed a list of nine viewpoints necessary to pass from the assumption of association to the one of causality (strength, consistency, specificity, temporality, biological gradient, plausibility, coherence, experimental evidence and analogy). There are, however, no absolute criteria for asserting the validity of scientific evidence in observational epidemiological studies. Therefore, Rothman ²² insists on the importance on relying on thorough criticism. Even experimental designs (randomised clinical trials) and meta-analysis are prone to bias that requires training and skills to recognise.

In our fictitious situation, probabilistic causality can estimate the attribution fraction of non-adherence to treatment in causing tachycardia. It can also estimate the risk of having an accident for patients with different degrees of cognitive disorders, or predict the probability of being hit by a car when riding a kick scooter out on the street.

Advantages: First, an epidemiological approach can constructs model of interaction and take dose-response relations into consideration. These models can also easily takes stochastic events into consideration and can provide a long-term prediction of probability of future events. Secondly, probability causality can also attribute different weights to multiple causes and include factors from different levels of organisation (e.g. biologic markers, behaviour traits and social indicators). Finally, probabilistic causality can easily consider the absence of a cause, as a cause (counterfactuals).²⁴

Limitations: The main limitation of epidemiology in science is that it does not provide any clear explanation of the underlying mechanisms linking a cause to an event. This approach is therefore entirely dependent on basic science and mechanical causality approaches to explain what it discovers. The second main difficulty is that probabilistic causality cannot distinguish simple associations from causal links unless we can assume that all other factors are held constant. Randomised clinical trials are therefore examples of controlled systems in which causal associations can be assumed. Conclusions can nevertheless be subject to confounding. Finally, for communication purposes, probabilistic causality concerns populations and not individuals. Patients often believe they are the exceptions and therefore do not always

PART  2  –  Causation  in  traffic  security  research  

rely on conclusions drawn from populations.

2.3.4 Causal modelling and psychology

Neuropsychological tests, behavioural questionnaires and psychological models are some of the instruments developed by causal modelling in psychology. In traffic medicine, these instruments are used to assess underlying brain functions by psychiatrists, psychologists, experts and primary care physicians.⁵ They are developed after observing multiple patients with similar symptoms or unusual behaviours. From these observations, traits can be attributed as parts of a given mental state.

Probabilistic causality is then used to assess the weight of each of these traits in defining the mental state. The final model’s ability to discriminate between patients with or without the mental state validates the entire procedure. This approach therefore also provides psychological explanations to certain mental states or behaviours.

Michon developed a model explaining that individuals accept a certain level of risk, from which they adapt their driving strategy.²⁵ This influences their ability to anticipate, plan, decide and react to situations. Young drivers are willing to take more risks than older adults. This could explain why speeding and alcohol causes so many accidents for young drivers. In our fictive situation, at a psychological level, this could also partially explain why the young man was ready to take the risk of riding a kick scooter on the sidewalk.

Advantages: First of all, causal modelling makes it possible to quantify an individual’s mental state. Secondly, causal modelling also makes it possible to study interactions between different mental states in a single individual or between individuals. Finally, psychological causal modelling also provides precious indications of the risk of certain behaviours by individuals.

Limitations: This method assumes that subpopulations share common underlying mechanisms explaining their behaviour. The way the population is classified is most often not, however, based on known underlying mechanisms. For similar mental states, models relying on different underlying concepts to classify disorders can therefore come up with different explanations. At a psychological level, however, it is not possible to test the validity of one model over another. Relying on lower-level studies remains essential to support causal modelling in psychology.

2.3.5 Statistical inference in social science

Social science relies on statistical inference to model a causal relationship and obtain the best “fit” between a constructed model and collected data.²⁶ Since the 80s, sociologists often rely on Mackie's theory of causality: the INUS-condition (insufficient but necessary parts of a condition which is itself unnecessary but sufficient). This approach uses a set of conditions and guarantees non-redundancy.

Approaches in sociology (INUS-condition, partial correlation and probabilistic cause) are nevertheless always based on correlation and association analysis. As such, the concept of causality in social science cannot be considered as valid. Sociology remains an interesting approach to identify associations between social indicators and societal events. However, causality remains difficult to accept at a social level.²⁶

2.3.6 Non-conformity causality

There are multiple examples of social laws regulating our behaviour. Let us focus on road safety regulations. If conditions necessary to cause an accident are in place, there is a given point in time when road users cannot change them anymore and the accident becomes inevitable. Traffic regulations have therefore been defined to help road users anticipate events and act accordingly with each other, while preserving traffic fluidity as much as possible. In other words, road regulations are meant to give people sufficient time to identify danger, react in consequence and prevent collisions.

These laws are governed by social agreement and conformity and as such are not deterministic. Nevertheless, not respecting such laws will modify constraints at a behavioural level. Road users will not be able to anticipate situations as expected.

Non-conformity causality can thereby be linked to mechanical causality at a lower level. For non-conformity causality to take place, it remains essential to verify that in the given circumstances, had the person conformed to social rules, the event would not have taken place. Only under this condition should we consider non-conformity causality as a potential cause of events.

The usual way to consider non-conformity causality is to consider all the sufficient conditions for an event to take place, and then estimate which is unexpected or unusual and define it as the cause. In our fictive situation, driving a car on the sidewalk while unconscious and riding a kick scooter on the sidewalk do not conform to social expectations and can be considered as causes of the accident. If it can be proven that the young man would not have been injured had he been on foot, or that the senior driver would not have lost control of his car had he not forgotten to take his medication, both the young man and the driver could be held responsible for the fractured femur.

Advantages: First, non-conformity causality can be studied by both probabilistic and mechanical approaches of causality. This approach can therefore provide explanations on the importance of societal rules in preventing or favouring events at a societal level. Secondly, this approach is the only approach that can attribute an individual’s level of responsibility to a given event. For this reason, it is often used by the legal system to establish links of causality between an individual’s action (or inaction) and damage.

Limitations: This type of cause is based on expectations and social conformity. The laws to which these causes respond are not absolute and can vary in time or space (different social groups can have different expectations and expectations can change over time). The constraint that this level has over other levels is therefore versatile, and transposing observations from one situation to another requires much caution.